Analysis of in situ hyperspectral data for nutrient estimation of giant sequoia

In this paper, we report some correlation analysis results between in situ hyperspectral data in the spectral range of approximately 350-900 nm and three foliage nutrient constituents. Two hundred and forty hyperspectral measurements were taken using a PSD1000 spectrometer at a giant sequoia plantation site in 1997. Foliage nutrient constituents (expressed in concentration, percentage of dry weight)-total nitrogen (TN), total phosphorus (TP) and total potassium (TK)-were measured from the same site. The potential of hyperspectral data for estimating foliage nutrient status was evaluated using univariate correlation and multivariate regression analysis methods with different types of predictors: original and the first-order derivative spectra, vegetation index (VI)-based, spectral position-based, area-based and principal component analysis (PCA)-based predictors. The eight VIs were constructed from the blue, green, red and near-infrared spectra bands; spectral position-based predictors consisted of parameters extracted from the blue, yellow and red edges, the green peak and the red well; area-based variables were calculated as the sum of the fit derivative values at each of the three edges; and the PCA-based predictors were obtained from principal component transformation applied to the first derivative spectra of the three edges. Results showed that the best foliage nutrient prediction was obtained with the PCA-based predictors in four-term prediction models for all three nutrient constituents. In univariate correlation analysis, it seems that only two VIs RB (R760-850/R350-400) and NRB ((R760-850-R350-400)/(R760-850+R350-400)) may be employed to predict all three nutrient constituents (p=0.01), and better R 2 values were obtained from the maximum first derivative spectra of blue edge and yellow edge and the spectral reflectance difference between green peak and red well for TN, TP and TK. The results obtained from the univariate analysis, however, indicate that the level of correlation between foliage nutrient concentrations and hyperspectral data is low. In fact for TN there existed rather low correlations with different types of variables. In general, the best univariate correlation results were obtained for TK, then TP, with TN the worst. For multiple regression with PCA-based predictors, the best results came from TN, then TK, while the worst was associated with TP.